Sakana AI Launches Fugu-Cyber Security Auditing Model

Tokyo-based startup Sakana AI has announced the release of 'Fugu-Cyber', a specialized cybersecurity model designed for automated network defense and software auditing. The model is integrated into Sakana's Fugu orchestration framework.

Fugu-Cyber is pre-trained on a massive dataset of known software vulnerabilities, exploit code, and system log files. The model outperformed generalized frontier models like GPT-4 on specialized penetration testing and security code review benchmarks.

Domain-Specific Models for Cyber Defense

The model is designed to assist security operations center (SOC) analysts by analyzing code repositories for security flaws in real time. It can automatically generate patch recommendations, reducing the time required to remediate zero-day vulnerabilities.

Automating Vulnerability Detection

Sakana AI has released Fugu-Cyber under a commercial API license, with an open-weights version available for academic research. The release highlights the growing trend of domain-specific AI models tailored for high-stakes security engineering.